DHCPv6 for IPv6 routing information, was Re: Mysterious missing DHCPv6 feature

[Benedikt Stockebrand]

Hi Bill and list,

bmanning at vacation.karoshi.com writes:

> 	what was hard to digest was the fact that the base IPv6 architeecture
> 	only works for about 85% of the existing use cases. That leaves about
> 	15% of existing, core use cases where the default IPv6 tool suite
> 	won't work as advertized or will screw up everything..

just out of curiosity---do you have any citable reference for those
numbers?

>       case in point,
> 	an internet exchange on shared media (yeah, these are increasingly rare
> 	but work w/ me here)  I've got a dozen routers on the same media, all
> 	running RA - telliing/shouting to all the other routers - NO, USE ME!

If you do the numbers, the overhead here is 2.2 bits/s/router
(advertising a single prefix, using Ethernet frames).  Even if you put
several hundred routers on a single link this should be negligible in
all but the most unusual cases.

More important, though: Why would anyone enable router discovery in an
all-router network?  All router implementations I've come across so
far treat router advertisements as a feature that can be enabled or
disabled independently of addresses or forwarding or whatever else.

> 	ND (neighboor discovery) has the same problems in a multi-homed network.

Sorry, but I can't follow you here.  What is so special about ND in a
multi-homed setup?

> 	It is for this reason, the other 15% that fall outside the "norm" that
> 	we need something like DHCP that is address family agnostic or I'm back to 
> 	hardcoding config data.

DHCP for legacy IPv4 has at least two serious legacy issues: Backward
compatibility with BOOTP and the need to start without a working
address.  Both of these issues are obviously irrelevant with IPv6, but
trying to be "address family agnostic" implies that the protocol you
think of will have to carry these legacy issues over to IPv6.

That said, the one design issue I have with DHCP on IPv6 is that I
still need relays in all client-inhabited subnets even for stateless
DHCP.  Maybe eventually somebody comes up with an updated spec that
allows clients to access ff05::1:3 (or another dedicated, routed MC
group) directly, but aside from that I am reasonably comfortable with
stateless DHCP the way it is.


Cheers,

    Benedikt

-- 
			 Business Grade IPv6
		    Consulting, Training, Projects

Benedikt Stockebrand, Dipl.-Inform.   http://www.benedikt-stockebrand.de/